Examples

Start with a routine workflow.

Good first candidates are repetitive, low-risk tasks with predictable outputs: support classification, extraction, tagging, summaries, and agent substeps.

Snippet

const response = await ai.chat.completions.create({
  model: "provider/model-large",
  messages: [
    { role: "system", content: "Classify the support request." },
    { role: "user", content: "I forgot my password and need a reset link." }
  ],
});

// MargIQ returns the normal provider response.
// The dashboard shows the requested model, selected model,
// savings estimate, routing source, and safety behavior.

Optimize

Support classification

Routine password reset tickets may earn a cheaper routing path.

Protect

Security escalations

Risky cases stay on the requested model as a safety guard.

Learn

Agent tool calls

Repeated substeps can become separate workflow profiles.